evening kid

I code too much

[evening kid @eveningkid]

If you wonder, this came up running qwen3.5-27b-64k locally!

[evening kid @eveningkid]

What does an LLM "think" about when you just say "hi"?

[evening kid @eveningkid]

I still have hope for this year's WWDC...but idk. Besides search bars at the bottom (sometimes - not even consistent), nothing has felt better

[evening kid @eveningkid]

Apple Liquid Glass was a bad redesign. Almost 1 year later, nothing about using it actually feels better. Even worse, there are still inconsistencies, feels less readable, wastes white spacing and doesn't even look better

[evening kid @eveningkid]

@tmikov Definitely. Problem is that code can often be right in multiple ways which opens infinite space for guesses… Even if cross-checked 10 times, at large it might still miss some of the “small stupid things”! Context management can be a big weakness too

[evening kid @eveningkid]

Fully committed

[evening kid @eveningkid]

@Zukoife @naval Exactly, most people already lack time to do what they want in their daily lives...

I think there might just be a small fraction of us, ready to spend time creating an app - which honestly can rarely be one-shotted

A business needing a tool vs the avg consumer

[Andrew Clark @acdlite]

How was work, Andrew?

[Christian Catalini @ccatalini]

1/ Some Simple Economics of AGI—🔥🧵
Right now, there is a low-grade panic running through the economy. Everyone is asking the same anxious question: what exactly is AI going to automate, and what will be left for us?

[evening kid @eveningkid]

@ccatalini @claudeai Great analysis and work, thank you.

I strongly believe we can mimic a human-like verification process - the same way you'd ask someone to verify whether x is true. AI can also determine when it can't physically or logically verify output/facts.

[evening kid @eveningkid]

@DilumSanjaya @Remotion @JNYBGR!! How cool!

[evening kid @eveningkid]

@tmikov Looks really nice, looking forward to the article!

[evening kid @eveningkid]

Whenever I see AI progress come just from ideas like this (not even new techniques), it proves that having the smartest people doesn’t always lead to the best results.

It’s more about thinking differently and finding new paths. Love that!

[Chris Laub @ChrisLaubAI]

Even more interesting:

Researchers found that lower WER doesn’t guarantee better retrieval.

Meaning: improving transcription accuracy hits diminishing returns.

Direct audio-to-retrieval learns what matters, not what was literally said.

[evening kid @eveningkid]

One thing AI has changed for me is that any task I used to do manually (just because I had to) now feels so tedious if I know AI can do it instead. Is this helping me or making me lazier?

[evening kid @eveningkid]

@DelphineBugner Or it read right through who we truly are... 😂

[evening kid @eveningkid]

@tmikov @davepl1968 Thank God my old printer is safe then! 🫡

[evening kid @eveningkid]

@tmikov @davepl1968 Name one thing that Windows does to make our lives easier...

[evening kid @eveningkid]

@sama GPT-5 often ends up in loops for me...makes it hard to use! Will keep trying.

[evening kid @eveningkid]

Interesting how context heuristics with Codex just don't work?

Left: gpt-5-codex MEDIUM, interrupted because it was "thinking" for >1min for a 1 liner. It had AUTO CONTEXT on...

Right: gpt-5-codex LOW, context manually added for the task, 10s.

We need better heuristics @sama!

[evening kid @eveningkid]

TIL NS prefixes in macOS development come from...NeXT OS, first released in the late 80s?!